From: aidotengineer
This article compiles insights from a survey of 100 AI Engineers during a “Frontier Feud” competition held in New York City [00:00:30], [00:01:07]. The survey aimed to gauge perspectives on various aspects of the AI industry.
AI Tools
When asked about AI tools that engineers love, the top answers included:
- Cursor (Number one answer) [00:18:04], [00:18:08], [00:20:43]
- Model APIs (Third answer) [00:20:04]
Model Considerations
Top considerations when choosing an AI model were identified as:
- Cost (Number one answer) [00:10:47], [00:10:51]
- Latency [00:11:04]
- Accuracy/Performance [00:11:28]
- Open Source vs. Closed Source [00:12:45]
Jobs at Risk from AI Disruption
Surveyed AI Engineers identified several jobs most at risk of AI disruption:
- Data Entry (Number one answer) [00:20:43]
- Software Engineers (Number three answer) [00:18:15], [00:18:17]
- Content Creation/Writing [00:20:10]
AI Industry Buzzwords
AI Engineers expressed fatigue with certain buzzwords, including:
- AGI (Artificial General Intelligence) (Number one answer) [00:15:30], [00:15:38]
- Agents [00:13:56]
- RAG (Retrieval-Augmented Generation) [00:16:21]
- Prompt Engineering [00:16:25]
- Co-pilots [00:15:12]
- Multimodality [00:14:56]
Influential AI Research and Papers
The most influential AI paper in history, according to the survey, is “Attention is All You Need” [00:17:49], [00:20:45].
Most influential AI researchers mentioned include:
- Geoffrey Hinton [00:06:21], [00:06:33]
- Yann LeCun [00:07:19], [00:07:23]
- Andrew Ng [00:09:09]
- Fei-Fei Li [00:09:13]
- Yoshua Bengio [00:09:19]
AI Engineer Nightmares
The biggest nightmare for an AI Engineer at 2 AM was identified as:
- An outage of the model or otherwise [00:20:52]
- Hardware failure / Info problem [00:17:53], [00:18:36]
Industry Benefiting Most from AI
The industry expected to benefit most from AI is Healthcare [00:20:57], [00:21:00].
Hot Takes on AI Development and Trends
Various “tech hot takes” shared during the event provide a glimpse into future AI engineering and trends:
- Future of Model Training: It’s predicted that at least one of the current major players training large models today will no longer be doing so by the end of the year [00:01:55].
- AI as Therapists: AI are suggested to make really good therapists [00:02:13].
- Non-AI Interrupted Thinking: Professionals in the AI industry should dedicate 20 minutes daily to paper-and-pencil thinking, uninterrupted by AI [00:02:27]. This is considered important for providing good content for AI models to be trained on, especially when AI models represent individuals [00:02:50].
- On-Device Models and Smaller Models: A significant trend towards the majority of deployed models being on-device within a year and a half, with smaller, hyperspecialized models orchestrated together, rather than solely relying on larger models and external data processing [00:03:31].
- AI Dating AI: In the future, dating apps might feature AIs dating each other [00:04:13].
- Beyond Transformers: Transformers are not seen as the final AI architecture, with eventual models being built using biological materials [00:04:19].
- Multi-Bot Conversations: One-on-one chatting with an AI bot is considered boring, suggesting that future conversations will involve at least two other bots in the room [00:04:37].
- “Token Wall”: The idea of hitting a “token wall” in AI development is dismissed as a “skill issue” [00:04:54].
Future of AI Engineering
An upcoming “State of AI Engineering” survey is being conducted to provide greater transparency in the industry regarding tools and workflows used by AI Engineers [00:21:39], [00:21:50].